肺癌始终位居肿瘤死亡率首位，超过80%的患者被确诊肺癌时已是中晚期,多模态影像是目前肺癌早期诊断的最佳检查方法,但是早期准确诊断肺小结节性质还不理想。本团队在十多年肺部轴位CT图像、三正交位CT图像恶性结节早期诊断的统计学方法研究基础上，本项目按照纳入排除标准入组300例肺部结节病患者，采集多模态PET/CT融合影像，首先研究非下采样双树复轮廓波变换方法，提取整体纹理、边缘细节及内部多层子代纹理信息；其次整合流行病学调查收集的个人基本信息、行为方式、环境因素、遗传规律、影像征象，以及大约1000个纹理信息指标，研究深度数据挖掘中的深度信念网络分类诊断模型新方法，构造不同基函数和组合基函数分类诊断模型；最后进行统计模拟与验证，编制智能化诊断系统程序，并请两名放射科主任医生进行盲法评价。为辅助放射科医生进行肺癌早期智能化诊断系统提供方法学支撑，为提高肺癌早期影像诊断准确率提供科学依据。

Lung cancer is always ranked first in cancer mortality, and more than 80% of patients diagnosed with lung cancer are in the middle or late stage. Multi-modal imaging is the best way to diagnose lung cancer early, but it is always difficult to differentiate benign and malignant pulmonary nodules. Based on the statistical study of the early diagnosis of malignant nodules of CT images and three orthogonal CT images, 300 cases with pulmonary nodules are enrolled according to the inclusion and exclusion criteria and multi-modal PET/CT fusion images are collected. Firstly, the non-sampling double-tree complex contour transform method is used to extract the whole texture information, the edge detail and the multi-layer internal sub-generation texture information. Secondly, integrating the basic information, Behavior, environmental factors, genetic rules, image signs through the epidemiological investigation, and no less than 1000 texture information indicators, to study the belief network of data mining for new diagnostic models to construct different basis functions and combined basis function classification diagnostic model ; Finally, statistical simulation and verification will be conducted, preparing for the programming of intelligent diagnostic system of lung cancer, and we will evaluate the system by inviting two senior titles of radiologists to make the diagnosis with blind method. The study aims to provide a scientific basis for assisting radiologists in intelligence and in similar research of early diagnosis of lung cancer, and to provide scientific basis for improving the accuracy of early diagnosis of lung cancer.